The host haloes of O I absorbers in the reionization epoch

Transcription

1 MNRAS 436, (2013) Advance Access publication 2013 October 1 doi: /mnras/stt1697 The host haloes of O I absorbers in the reionization epoch Kristian Finlator, 1 Joseph A. Muñoz, 2 B. D. Oppenheimer, 3,4 S. Peng Oh, 5 Feryal Özel 6 and Romeel Davé 6,7,8,9 1 Dark Cosmology Centre, Niels Bohr Institute, Copenhagen University, Juliane Maries Vej 30, DK-2100 Copenhagen O, Denmark 2 University of California Los Angeles, Department of Physics and Astronomy, CA 90095, USA 3 Leiden Observatory, Leiden University, PO Box 9513, Leiden, the Netherlands 4 CASA, Department of Astrophysical and Planetary Sciences, University of Colorado, 389-UCB, Boulder, CO 80309, USA 5 Department of Physics, University of California Santa Barbara, Santa Barbara, CA 93106, USA 6 Astronomy Department, University of Arizona, Tucson, AZ 85721, USA 7 University of the Western Cape, Bellville, Cape Town 7535, South Africa 8 South African Astronomical Observatories, Observatory, Cape Town 7525, South Africa 9 African Institute for Mathematical Sciences, Muizenberg, Cape Town 7545, South Africa Accepted 2013 September 5. Received 2013 August 30; in original form 2013 May 17 ABSTRACT We use a radiation hydrodynamic simulation of the hydrogen reionization epoch to study O I absorbers at z 6. The intergalactic medium (IGM) is reionized before it is enriched; hence, O I absorption originates within dark matter haloes. The predicted abundance of O I absorbers is in reasonable agreement with observations. At z = 10, 70 per cent of sightlines through atomically cooled haloes encounter a visible (N O I > cm 2 ) column. Reionization ionizes and removes gas from haloes less massive than M, but 20 per cent of sightlines through more massive haloes encounter visible columns even at z = 5. The mass scale of absorber host haloes is times smaller than the haloes of Lyman-break galaxies and Lyman α emitters, hence absorption probes the dominant ionizing sources more directly. O I absorbers have neutral hydrogen columns of cm 2, suggesting a close resemblance between objects selected in O I and H I absorption. Finally, the absorption in the foreground of the z = quasar ULAS J cannot originate in a dark matter halo because halo gas at the observed H I column density is enriched enough to violate the upper limits on the O I column. By contrast, gas at less than one-third the cosmic mean density satisfies the constraints. Hence, the foreground absorption likely originates in the IGM. Key words: galaxies: evolution galaxies: formation galaxies: haloes galaxies: highredshift quasars: absorption lines cosmology: theory. 1 INTRODUCTION Mapping out the progress of hydrogen reionization and understanding the nature of the sources that drove it constitute two of the central challenges that astronomy will confront over the coming decade (National Research Council 2010). The cosmic microwave background (CMB) constrains reionization to be roughly 50 per cent complete at some point between z = 9 and 11.8, although the results depend on the shape of the assumed reionization history (Pandolfi et al. 2011; Mitra, Choudhury & Ferrara 2012; Hinshaw et al. 2013). The classic approach of measuring the neutral hydrogen fraction directly from the Lyman α (Lyα) forest becomes increasingly difficult at redshifts beyond z = 6 owing to the fact that Lyα absorption kfinlator@dark-cosmology.dk saturates for neutral hydrogen fractions in excess of 10 3 (Fan et al. 2002). In response to this challenge, a number of alternative techniques have been developed involving the abundance of Lyα emitters (Ouchi et al. 2010; Treu et al. 2012) or Lyman-break galaxies (Muñoz & Loeb 2011; Finkelstein et al. 2012; Oesch et al. 2013; Robertson et al. 2013), the statistics of dark pixels or gaps in the Lyα forest (Mesinger 2010; McGreer, Mesinger & Fan 2011), or the presence of damping wings in quasar spectra (Bolton et al. 2011; Schroeder, Mesinger & Haiman 2013). Each of these approaches combines unique strengths and weaknesses, hence it is necessary to consider a diverse variety of approaches together in order to overcome the weaknesses of any individual one. One probe that has received relatively little attention involves the study of low-ionization metal absorbers (Oh 2002; Furlanetto & Loeb 2003). If diffuse regions of the pre-reionization intergalactic medium (IGM) were enriched with metals whose ionization C 2013 The Authors Published by Oxford University Press on behalf of the Royal Astronomical Society

2 O I-absorbing haloes 1819 potential is similar to that of hydrogen, then it may be possible to measure the ionization state of the metals directly and use this to trace the ionization state of the IGM as a whole. Recently, Becker et al. (2011) searched for low-ionization metal absorbers in moderate- and high-resolution spectra of 17 quasars at redshifts They found that the abundance of systems at z 6 roughly matches the combined number density of damped Lyα systems (DLAs; <N H I /cm 2 ) and sub-dlas (10 19 < N H I /cm 2 < )atz 3. Furthermore, the velocity widths of the high-redshift absorbers are similar to those of the DLAs, although with weaker equivalent widths. The authors concluded that low-ionization metal absorbers trace low-mass haloes rather than neutral regions in the diffuse IGM. Modelling O I absorbers in order to study the viability of this scenario requires a model that treats the inhomogeneous ionization and metal enrichment fields simultaneously. In Oppenheimer, Davé & Finlator (2009), we used a cosmological hydrodynamic simulation that assumed a spatially homogeneous extragalactic ultraviolet ionizing background (EUVB) to study metal absorbers in the reionization epoch. The ionization field was adjusted in post-processing to consider scenarios in which there was no EUVB, a spatially homogeneous EUVB, and an inhomogeneous model in which the EUVB at any point was dominated by the nearest galaxy. It was found that the O I absorber abundance was dramatically overproduced in the absence of an EUVB, and underproduced under the assumption of an optically thin EUVB or a simple model in which the EUVB at any point was governed by the nearest galaxy. This work neglected two important aspects of the radiation field. First, the clustered nature of ionizing sources means that the EUVB at any point is determined by the combined influence of many galaxies rather than just the nearest one (Barkana & Loeb 2004; Furlanetto, Zaldarriaga & Hernquist 2004a,b; Furlanetto & Oh 2005). Secondly, dense sources acquire a multiphase ionization structure consisting of an optically thick core and an optically thin atmosphere. Modelling the ionization front that separates these phases requires a spatial resolution of 1 physical kpc (Schaye 2001; Gnedin & Fan 2006; McQuinn, Oh & Faucher-Giguère 2011), which was not achievable through the simple treatment adopted in Oppenheimer et al. (2009). For these reasons, the spatial dependence of the assumed radiation field was incorrect. Hence, while our previous study confirmed that there is enough oxygen to account for observations, the crude treatment of the EUVB meant that direct comparison with observations was preliminary. Here, we remedy these deficiencies by studying the nature of O I absorption using cosmological simulations in which the EUVB and the galaxies are modelled simultaneously and self-consistently. We focus on O I absorbers because the abundance of oxygen leads to high O I columns while the proximity of its ionization potential to that of hydrogen means that the neutral oxygen fraction can be obtained trivially from the neutral hydrogen fraction. The goals of the current study are: (1) to study the relative spatial distributions of enriched and ionized gas and determine which portion of the IGM O I observations likely probe; (2) to understand the impact of reionization on the sources of O I absorption; (3) to compare the predicted and observed abundances of O I absorbers; and (4) to compare the H I and O I absorption properties of halo gas in the reionization epoch. Additionally, we will use our model to interpret observational constraints on the abundance of O I in the absorbing system that lies in the foreground of the z = quasar ULAS J (Mortlock et al. 2011). In Section 2, we introduce our simulations. In Section 3, we explore the spatially inhomogeneous ionization and chemical enrichment fields in our simulations. In Section 4, we use insights from our simulations to model the abundance of neutral oxygen absorbers as a function of redshift and compare with observations. We also compare the predicted H I and O I absorption properties of reionization-epoch haloes. In Section 5, we discuss our results with an eye towards future modelling efforts and in Section 6 we summarize. 2 SIMULATING REIONIZATION AND ENRICHED OUTFLOWS 2.1 Simulations We use hydrodynamic simulations to model the inhomogeneous ionization and metallicity fields. These simulations are built on the parallel N-body + smoothed particle hydrodynamics (SPH) code GADGET-2 (Springel 2005) and include treatments for radiative cooling, star formation and momentum-driven galactic outflows (except for one simulation as we describe below). We model the EUVB on the fly by solving the moments of the radiation transport equation on a Cartesian grid that is superposed on our simulation volume. The ionizing emissivity within each cell is determined by the local star formation rate (SFR) density, with a metallicity weighting based on the stellar population models of Schaerer (2003). The fraction of ionizing photons that escape into the IGM varies depending on the simulation (see below). The radiation and ionization fields are updated simultaneously using an iterative procedure. For details on all of these ingredients, see Finlator, Davé&Özel (2011b) and Finlator et al. (2012). Three of the four simulations account for the ability for dense gas to acquire an optically thick core on spatial scales beneath the resolution limit of our radiation transport solver. We introduced this subgrid treatment in Finlator et al. (2012), but we review it here as it is a critical ingredient for modelling low-ionization metal absorbers. Directly resolving the ionization fronts that isolate optically thick regions requires a spatial resolution of 1 physical kpc (Schaye 2001; Gnedin & Fan 2006; McQuinn et al. 2011). By contrast, our highest resolution simulation discretizes the radiation field using mesh cells that are h 1 kpc wide (comoving). While this allows us to model our volume s reionization history with 10 5 cells, the resolution remains roughly a factor of 10 too coarse to resolve Lyman limit systems (LLS; N H I > cm 2 ). We overcome this limitation through a generalization of the Haehnelt, Steinmetz & Rauch (1998) self-shielding scenario. Each SPH particle is exposed to an EUVB that is attenuated by an optical depth τ Ɣ that varies with the local overdensity ρ/ ρ as τ Ɣ = ( / lls ) b. The characteristic scale lls is the overdensity of systems through which an optical depth of unity is expected under the assumption that the gas is in hydrostatic equilibrium. It depends on the local temperature, redshift, and the amplitude of the EUVB (Schaye 2001), and it grows from 10 at z = 10 to 100 by z = 6 (fig. 2 of Finlator et al. 2012). We set the power-law slope b = 3, although this choice does not affect the results significantly. We also add the opacity of the self-shielded gas to the overall opacity field for self-consistency. Gas with < lls sees an unattenuated EUVB. This treatment yields an ionization field in which gas that is more than a few times more dense than lls is neutral, in agreement with simulations that model the ionization field with higher resolution in a post-processing step (McQuinn et al. 2011). Table 1 shows our suite of simulations. The naming convention encodes the simulation parameters. For example, the r6n256wwwrt16d simulation subtends 6 h 1 Mpc (r6) using

3 1820 K. Finlator et al. Table 1. Our simulations. The fiducial simulation is indicated in bold. Name L a RT grid Outflows? Self-shielding? r6n256wwwrt16d Yes Yes r6n256nwwrt16d No Yes r9n384wwwrt48d Yes Yes r6n256wwwrt Yes No a In comoving h 1 Mpc particles (n256) with outflows (ww) and discretizes the radiation field using 16 3 cells (wrt16) including subgrid selfshielding (d). For all but the r6n256wwwrt simulation, the ionizing escape fraction varies with redshift as { ( 1+z ) κ fesc,5 z<10 6 f esc = (1) 1.0 z 10. Here, the normalization f esc, 5 sets the escape fraction at z = 5, which we tune to match the observed ionizing emissivity at that redshift (Kuhlen & Faucher-Giguère 2012). The slope κ controls how strongly f esc varies with redshift and is tuned to reach 1 at z = 10. These requirements lead us to adopt f esc, 5 = and κ = 4.8 for the r6n256wwwrt16d and r9n384wwrt48d simulations. The r6n256nwwrtd simulation is similar but does not include outflows. Without outflows, the predicted SFR density is higher, hence we require a lower escape fraction in order to match observations; we adopt f esc, 5 = and κ = The r6n256wwwrt simulation does not include self-shielding and assumes a constant ionizing escape fraction f esc = 0.5. Note that our r9n384wwwrt48d run includes the same underlying physics as the r6n256wwwrt16d run but times more volume and a finer radiation transport mesh, giving it the highest dynamic range that we have modelled to date. It required CPU hours on 128 processors to reach z = 6. It is the fiducial simulation volume for the current study. All simulations incorporate the same resolution such that the mass of a halo with 100 dark matter and SPH particles is M, and the gravitational softening length is 0.1 kpc (Plummer equivalent; proper units at z = 6). We generate the initial density field using an Eisenstein & Hu (1999) power spectrum at redshifts of 249 and 200 for simulations subtending 6 and 9 h 1 Mpc, respectively. We initialize the IGM temperature and neutral hydrogen fraction to the values appropriate for each simulation s initial redshift as computed by RECFAST (Wong, Moss & Scott 2008), and we assume that helium is initially completely neutral. All simulations assume a cosmology in which M = 0.28, = 0.72, b = 0.046, h = 0.7, σ 8 = 0.82, and the index of the primordial power spectrum n = The focus of our current work is the spatial distribution of neutral oxygen. Our simulations do not evolve the ionization state of oxygen on the fly because it contributes negligibly to the total opacity. In order to compute the abundance of neutral oxygen, we combine in post-processing the predicted neutral hydrogen fraction and total oxygen abundance (which are both modelled on the fly) with the assumption that hydrogen and oxygen are in charge exchange equilibrium at the local gas temperature. To do this, we use the expression (Oh 2002) N O I N H I = 9 exp N O II 8 N H II ( E k B T ), where E = 0.19 ev is the difference between the first ionization potentials of oxygen and hydrogen and T is the local temperature. 2.2 Comparison to observed reionization history A challenge to modelling reionization involves the problem of creating a high enough ionizing emissivity at early times to match the observed optical depth to Thomson scattering in the CMB τ es without overproducing the observed amplitude of the EUVB after z = 6. Models that assume that a constant fraction f esc of all ionizing photons escape into the IGM can match one, but not both of these constraints (Finlator et al. 2011b). Observations can be reconciled by assuming that f esc varies with either halo mass (Yajima, Choi & Nagamine 2011; Alvarez, Finlator & Trenti 2012) or redshift (Kuhlen & Faucher-Giguère 2012; Mitra, Ferrara & Choudhury 2013). Our fiducial simulation uses a time-dependent f esc to overcome this problem (Section 2.1). Here we briefly discuss how well it matches observational constraints. If we assume that helium is singly ionized with the same neutral fraction as hydrogen for z>3and doubly ionized at lower redshifts, then our r9n384wwwrt48d simulation yields an integrated optical depth of τ es = This falls within the observed 68 per cent confidence interval of ± (Hinshaw et al. 2013), indicating that reionization is sufficiently extended. 1 The predicted optical depth in the Lyα transition at z = 6 is 2.6. As before, this is somewhat lower than the observed lower limit (>5; Fan et al. 2006), implying that the predicted radiation field is slightly too strong. If true, then our simulations could underestimate the abundance of O I absorbers at z = 6. However, we note that our model is not unique in failing to reproduce the weak radiation field observed at z = 6. In particular, observations suggest that the ionizing emissivity strengthens from <2.6 to 4.3 ± 2.6 (in units of s 1 Mpc 3 ) from z = 6 to 5 (Kuhlen & Faucher-Giguère 2012); such rapid growth is quite difficult to accommodate within a model where f esc varies smoothly with redshift (see, however, Alvarez et al. 2012). For redshifts below z = 6, we use predictions from the r6n256wwwrt16d run, which incorporates the same physical treatments as the fiducial simulation but subtends a smaller volume. At z = 5, this simulation yields an effective optical depth to Lyα absorption of τ α = 3.1, marginally consistent with the observed range of 2 3 (Fan et al. 2006). In summary, the assumption of an evolving escape fraction allows our simulations to match the observed τ es while only weakly conflicting with constraints on the post-reionization EUVB. Hence, the predicted IGM ionization structure, thermal history and the star formation history are plausible starting points for studying lowionization metal absorbers during the reionization epoch. In this work, we will show that they primarily trace star formation in lowmass haloes and use their predicted abundance as a new test of the model. 2.3 The importance of self-shielding Having introduced our simulations, we are now in a position to demonstrate the importance of self-shielding. We compare in Fig. 1 the mean radial density profiles of all oxygen (solid) and neutral oxygen (dashed) in simulations without (light blue) and with (heavy red) self-shielding (the r6n256wwwrt and r6n256wwwrt16d simulations, respectively). We produce these curves by averaging over 1 In Finlator et al. (2012), we noted that the predicted τ es of underproduced the observations reported in Komatsu et al. (2011). The current agreement results from the fact that measurements of small-scale anisotropy in the CMB have since brought the inferred τ es down (Story et al. 2012; Hinshaw et al. 2013). Considering broader classes of reionization histories also decreases the inferred τ es (Pandolfi et al. 2011).

4 O I-absorbing haloes 1821 Figure 1. The radial density profiles of all oxygen (solid) and neutral oxygen (dashed) in haloes of mass M in simulations without (blue) and with (red) self-shielding. Gas associated with galaxies has been removed. The right vertical dashed segment indicates the virial radius. Self-shielding enhances the neutral oxygen abundance significantly at all radii. haloes in bins of mass and radius; see Section 3.2 for details. The solid curves overlap, indicating that simulations with similar reionization histories and identical models for galactic outflows yield similar metal density profiles. By contrast, the light blue dashed curve lies nearly a factor of 10 below the heavy red dashed curve, indicating that the neutral oxygen abundance is artificially underestimated by a factor of 10 if self-shielding is ignored. It is interesting to note that our previous simulations underpredicted the observed abundance of O I at z = 6 by a factor of 15 (fig. 11 of Oppenheimer et al. 2009), independent of whether the ionization state was modelled using a spatially homogeneous EUVB or a background dominated by the nearest galaxy. In that work, the offset was interpreted as evidence for a partially neutral universe at z = 6. By contrast, Fig. 1 suggests that the disagreement may owe to the absence of self-shielding in that work. If so, then O I observations may indeed be consistent with a reionized universe at z = 6, with the observed systems arising entirely in optically thick regions such as galaxies. Our new simulations enable us to explore this possibility. 3 METAL ENRICHMENT AND IONIZATION 3.1 The competition between enrichment and reionization Early interest in low-ionization metal absorbers centred on the possibility that the diffuse IGM could be enriched before it was reionized (Oh 2002; Furlanetto & Loeb 2003). The question of whether this works can be distilled to a competition between the growth of enriched regions and the growth of ionized regions. If galaxies reionize their environments more quickly than they enrich them, then O I absorption will be dominated by self-shielded clumps rather than by low-density regions that have not yet been reionized. On the other hand, if galactic outflows enrich the diffuse IGM (that Figure 2. The relationship between metallicity, overdensity, neutral fraction and reionization at z = 10 (upper panel) and z = 6 (lower panel). The blue dashed curves show the mean trend of metallicity versus overdensity while blue dashed error bars enclose the middle 50 per cent wherever the median is non-zero. Light, medium and heavy black contours represent neutral hydrogen fractions of 10 5,10 2 and 0.5, respectively. The volumeaveraged neutral fractions at z = 6 and 10 are and 0.83, respectively. The red dotted curves indicate the minimum metal mass fraction to produce an observable absorber for a hydrostatically bound region at 10 4 K. is, regions with overdensity ρ/ ρ < 10) very quickly, then there may be a substantial reservoir of neutral metals that can be observed in absorption prior to the completion of reionization. This idea seems unlikely at a glance because a galaxy s ionization front ought to grow more rapidly than its metal pollution front. However, ionizing sources are not necessarily time steady, and if star formation is bursty then the IGM surrounding a galaxy can recombine once its OB stars evolve off the main sequence. The metals ejected into the IGM are permanent, however, and could become visible in low-ionization transitions (Oh 2002). In order to motivate a detailed study of how this competition unfolds, we show in Fig. 2 the relationship between overdensity, metallicity, and neutral hydrogen fraction before and after the completion of reionization. The blue dashed curves show the massweighted mean metal mass fraction as a function of overdensity. As was seen in fig. 4 of Oppenheimer et al. (2009), the mean metallicity grows significantly in regions that are moderately overdense (ρ/ ρ < 100) while in denser regions it rapidly reaches an equilibrium value that is driven by self-regulated star-forming regions (Finlator & Davé 2008). Importantly, outflows give rise to a reservoir of enriched gas at overdensities of even at z = 10. The red dotted curves show the minimum metal mass fraction for neutral regions in hydrostatic equilibrium at a temperature of 10 4 Kto produce an O I column greater than cm 2 as a function of overdensity. Comparing the red dotted and blue dashed curves indicates that overdense regions would produce observable absorption if they were homogeneously enriched to the mean level and neutral. In order to ask whether the enriched regions could be neutral, we use contours to show the neutral hydrogen fraction as a function of density and metallicity. The heaviest or innermost contours illustrate

5 1822 K. Finlator et al. the phase space where the neutral hydrogen fraction is 50 per cent, hence they mark the transition from diffuse, ionized gas to condensed, neutral gas. The low-density limit of this region lies near the mean density at z = 10, implying that much of the metal mass that is expelled into the IGM may remain neutral. Even at z = 6, the bulk of the gas in the Lyα forest ( 10) is on average neutral and enriched, implying the presence of a substantial forest of low-ionization metal absorbers. Fig. 2 seems to support the use of O I to probe the progress of reionization, but this could be misleading. The crucial question is whether the enriched regions are neutral and vice versa. For example, a small population of enriched, ionized lumps could drive up the mean metallicity without suppressing the mean neutral fraction. To amplify this possibility, we use blue dashed error bars enclose the middle 50 per cent of metallicities wherever the median metallicity is non-zero. They agree with the mean for overdensities above 30, but at lower densities the median vanishes, indicating that the mean is driven by a small set of enriched regions. The need for detailed study of the IGM phase structure is further emphasized by observational evidence that metals mix quite poorly with the ambient IGM (Schaye, Carswell & Kim 2007). If the ionization state is similarly inhomogeneous, then the heavy averaging inherent in Fig. 2 could be quite misleading. Our simulations model the inhomogeneous ionization and metallicity fields directly (subject to resolution limitations as described in Section 2.1), allowing us to address these questions. In order to gain intuition into where O I absorbers live with respect to dark matter haloes and LLSs, we show in Fig. 3 maps of (top to bottom) gas density, temperature, metallicity, H I column and O I column for four different dark matter haloes at two different redshifts. The left two columns show how, at z = 10, much of the volume is filled with neutral hydrogen as expected for a universe that is only 50 per cent ionized. Near haloes, this enriched gas produces O I columns stronger than cm 2 well outside of the virial radius. By z = 6, the gas around similarly massive systems (right two columns) is even more enriched, but by now the ionization fronts have penetrated deeper into the halo, ionizing much of the diffuse gas that would have been visible as low-ionization absorbers at z = 10. Countering this trend is the growing abundance of satellite haloes, the cores of which are neutral and enriched. As a result, low-ionization absorbers are common around haloes at both z = 10 and 6. Fig. 3 strongly suggests that O I absorbers trace enriched gas within dark matter haloes rather than the diffuse IGM. A more quantitative way to ask which regions contain gas that is both enriched and neutral enough to yield observable absorption is to compute the characteristic column density as a function of density. If a parcel of gas is in hydrostatic equilibrium, then its characteristic O I column density N O I,c is Z O n O I N O I,c L J ρ b, (2) m O n O where L J is the Jeans length, ρ b is the mass density in baryons, Z O is the mass fraction in oxygen, m O is the mass of an oxygen atom and n O I /n O is the neutral oxygen fraction (see equations 3 and 4 of Schaye 2001). We compute the characteristic column density for each overdense particle using the local density, temperature, metallicity and ionization state, and show the resulting trends at two representative redshifts in Fig. 4. The dashed horizontal line shows the current 50 per cent observational completeness limit for selecting absorbers in O I (Becker et al. 2011). Gas at the mean density (ρ/ ρ 1) is ionized by the nascent EUVB even at z = 10, hence it does not produce observable O I absorption. While we cannot apply equation (2) to underdense gas because it is not expected to be in hydrostatic equilibrium (Schaye 2001), the trend in Fig. 4 strongly suggests that it does not produce visible absorption either. At higher densities, the threshold for gas to be optically thick and hence neutral grows from 20 at z = 10 to >300 at z = 6. Given that gas with overdensity greater than 10 is predicted to be enriched (Fig. 2), the evolving threshold for it to be optically thick is also the threshold for it to produce visible O I absorption. In summary, our simulations predict that ionization fronts precede metal pollution fronts, and that regions, once ionized, remain ionized. This owes partially to the fact that hydrogen-cooling haloes produce stars steadily until their environments are reionized (note that Wise & Abel 2008 find that star formation becomes a steadystate process in pre-reionization haloes more massive than 10 7 M, an order of magnitude below our resolution limit) and partially to the clustered nature of galaxy formation, although a detailed analysis of the relative roles of these factors is currently impossible owing to our small volumes. Consequently, diffuse gas does not produce observable absorption in low-ionization transitions. For the rest of this work, we will therefore focus on low-ionization metal absorption that occurs within dark matter haloes. 3.2 Radial profiles In this section, we explore how the radial density profiles of gas, total metals and neutral metals vary with mass and redshift. We will consider haloes that are both more and less massive than 10 9 M because this marks the approximate threshold above which haloes can accrete gas even in the presence of an EUVB. For consistency with Finlator et al. (2011b), we will refer to the lower mass haloes as photosensitive and the more massive haloes as photoresistant. We compute radial density profiles by stacking haloes in bins of mass and averaging within each radial bin. By computing the density of O I within each shell directly (rather than computing the oxygen density and neutral fractions and multiplying them), we preserve small-scale inhomogeneities in the metallicity and enrichment fields. As a demonstration of how our spherically averaged radial profiling works, we show in Fig. 5 the density, neutral fraction and column density profiles for our most massive halo at z = 10 (left-hand column in Fig. 3). The solid blue curve shows that the halo possesses an enriched neutral core that is associated with O I column densities above cm 2 out to at least 0.2 virial radii (R vir = 6.6 kpc; bottom panel). This is dominated by star-forming gas in the central galaxy. Outside of this core there is an enriched, partially neutral reservoir that generates observable column densities (N O I > cm 2 ;the black dot dashed line in the bottom panel) out to the virial radius. Our approach works well if the gas is distributed spherically symmetrically, but it breaks down if the majority of a halo s gas is bound into a small number of satellite systems because the geometric cross-section for a sightline to intersect a satellite is smaller (and the associated gas column higher) than if the satellite s gas was distributed in a shell. Additionally, the fact that our simulations neglect ionizations owing to the local radiation field means that the abundance of neutral oxygen within galaxies could be overestimated (we will return to this point in Section 5). In order to mitigate these problems, we use SKID 2 to identify and remove all gas that is associated with galaxies before computing density profiles. 2

6 O I-absorbing haloes 1823 Figure 3. Maps of gas density, temperature, metallicity, H I column, O I column and dark matter density for two halo masses at z = 10 and 6. Each panel spans 100 proper kpc, and the circles indicate the virial radii of the parent haloes. At z = 10, the weak EUVB leaves an abundant population of LLS, but only those that lie near haloes are associated with a significant O I column. By z = 6, O I absorbers have retreated well into the central halo s virial radius and are of generally higher column density.

7 1824 K. Finlator et al. Figure 4. The characteristic column density for O I absorption as a function of overdensity in regions with non-zero metallicity. Contours enclose 67 and 99 per cent of gas particles at z = 10 (top) and 6 (bottom). The dashed line indicates the 50 per cent completeness limit (Becker et al. 2011). Regions with overdensities of less than 10 are never visible in absorption for z<10. Figure 6. The radial profiles of hydrogen density, oxygen mass fraction, neutral fraction and neutral oxygen density in a M halo at z = 6 in simulations without outflows (magenta dashed) and with outflows (black solid). The virial radius is 7.3 kpc. The red horizontal long dashed line in the top panel indicates the threshold density for forming stars. The red vertical long dashed line indicates the gravitational softening length. Outflows dominate the CGM at small radii and generate an atmosphere of ionized, enriched gas at large radii. They do not enhance the geometric cross-section for absorption in low-ionization transitions. Figure 5. Sample profiles of O I density n O I, neutral oxygen fraction X O I and neutral oxygen column density N O I as a function of radius for the M central halo at z = 10 shown in the left-hand column of Fig. 3. The solid blue profiles include both interstellar and intergalactic gas, while the dashed red profiles exclude all gas that is bound within resolved galaxies. The n O I and X O I profiles are both smoothed with a 1 kpc boxcar filter. The black dot dashed curve in the bottom panel indicates current observational limits (Becker et al. 2011). We do not trace the profiles beyond a virial radius, hence they vanish there artificially. Excluding ISM gas suppresses the O I column at small radii owing to the central galaxy and at large radii owing to satellites, but on the whole the halo remains observable out to the virial radius. The dashed red curve shows the same density profile as the solid blue curve, but without galaxy gas. This step suppresses the density of neutral gas significantly near the halo s core, but at larger radii the difference is slight because the gas in resolved satellites is subdominant to the combined contributions of unresolved satellites and the circumgalactic medium (CGM). Note that the column densities in the bottom panel are notional because they are derived from spherically averaged profiles. In the second part of this work, we will relax the assumption of spherical symmetry and use a ray-casting approach to compute the geometric absorption cross-section, enabling a more accurate comparison with observations. Having demonstrated how we compute spherically averaged profiles, we now ask how outflows impact the CGM. We show in Fig. 6 the radial density profiles of gas and metals in M haloes in simulations without (dashed magenta) and with (solid black) galactic outflows. Panel (a) compares the gas densities. The profiles flatten below 1 kpc because gas at these radii is dense enough to support star formation, which suppresses the gas density. Recalling that we have removed galactic gas from these profiles, we see that there is no circumgalactic gas within 0.1R vir unless outflows put it there because inflows at these radii collapse quickly on to the central galaxy. At larger radii, the profiles are nearly coincident because most of the gas is infalling rather than outflowing. Panel (b) shows the oxygen metallicity profile. Near the central star-forming region (within 2 kpc), outflows give rise to an enriched atmosphere. At larger radii, simulations without outflows still suggest an enriched CGM. However, outflows clearly boost the mean metallicity beyond 0.2R vir (see also Oppenheimer et al. 2009).

8 O I-absorbing haloes 1825 We show the neutral oxygen fraction in panel (c). Nearly all of the CGM s metals are neutral in the absence of outflows. These metals could correspond either to star-forming gas in satellite haloes that are too small to be identified and removed by our group finder, or to moderately enriched inflowing streams; simulations with higher resolution would be required to distinguish between these possibilities. By contrast, the neutral metal fraction drops at large radii in simulations with outflows. This does not owe to differences in the EUVB because f esc is tuned separately for each simulation to produce similar EUVBs by z = 5. Instead, it indicates that outflows tend to be highly ionized. A detailed analysis of the thermal structure of outflows is beyond the scope of the present work, but for reference we note that, for gas particles that have recently been ejected at z = 7, our model predicts a median density of 0.4 times the mean baryon density and a median temperature of K. For such gas, the recombination time exceeds a Hubble time, hence it is expected to be largely ionized by z = 7. The product of the curves in the top three panels is proportional to the neutral oxygen density, which we show in panel (d). This panel confirms that metals that are ejected in outflows are generally ionized and do not enhance the probability that the host halo will be observable as a low-ionization metal absorber. They must instead be sought using high-ionization transitions such as C IV (Oppenheimer &Davé 2006; Borthakur et al. 2013) or O VI (Tumlinson et al. 2011). Note that this conclusion is not necessarily general. For example, Ford et al. (2013) have shown that outflows enhance the abundance of Mg II absorbers around M haloes at low redshifts (their fig. 14). In Fig. 7, we evaluate how the O I density profile varies with mass prior to the completion of reionization. Examining the photosensitive haloes first (panel a), we find that the central star-forming region (<0.5R vir ) contains a significant reservoir of metals because Figure 7. The radial profiles of oxygen density in haloes of mass log 10 (M h /M ) = 8.3 (panel a) and 9.5 (panel c) at z = 10 in our fiducial simulation. Panels (b) and (d) show the corresponding neutral fractions. Profiles are smoothed with a 0.3 kpc boxcar filter for clarity. At z = 10, CGM metals are completely neutral in photosensitive haloes and mostly neutral in photoresistant haloes. Figure 8. The same as Fig. 7 but at z = 6. We also distinguish total and neutral oxygen density in panels (a) and (c) as indicated. Once reionization completes, the EUVB penetrates to roughly 0.5R vir in both photosensitive and photoresistant haloes. A tail of partially neutral gas extends to the virial radius in the photoresistant haloes, suggesting that they could dominate O I absorption statistics once reionization is complete. these haloes are massive enough to cool their gas and form stars. Furthermore, panel (b) shows that their metals remain completely neutral out to the virial radius because they inhabit preferentially underdense regions where the EUVB remains weak at z = 10. We will show in Sections that these haloes have a geometric cross-section to absorption that is not small compared to the halo cross-section and that, consequently, they dominate O I absorption statistics prior to the completion of reionization. Turning to photoresistant haloes, we see that the density of metals is one to two orders of magnitude higher than in the photosensitive haloes owing to their higher star formation efficiencies. The neutral fraction drops below unity outside of roughly 0.6R vir because these haloes inhabit preferentially dense regions where the EUVB takes hold at earlier times. Even at the virial radius, however, the neutral fraction exceeds 50 per cent, suggesting that these haloes generate high-column absorbers even though they are subject to a stronger EUVB. As the EUVB strengthens and ionization fronts penetrate the CGM, we expect the O I density profiles to evolve. We show in Fig. 8 how the profiles in the same mass ranges have evolved by z = 6 (note that the virial radii are also larger now). Photosensitive haloes have grown a substantially higher total oxygen density, particularly near their cores ( 0.3R vir ). These haloes are able to continue forming new stars and metals even at z = 6 owing to the fact that gas that cooled prior to reionization remains bound and star forming for several dynamical times following overlap (Dijkstra et al. 2004). However, the gas is only neutral within 0.5R vir. Gas at larger radii is completely ionized by the EUVB. In fact, our simulations suggest that haloes near the hydrogen-cooling limit ( 10 8 M ) are evaporated by the EUVB in a process similar to the evaporation of minihalo gas (Shapiro, Iliev & Raga 2004). For both of these reasons, the abundance of neutral oxygen at the virial radius of

9 1826 K. Finlator et al. photosensitive haloes declines and their contribution to lowionization metal absorbers diminishes as reionization proceeds. Photoresistant haloes are also more ionized than at z = 10 (panels c and d), but the effect is weaker than for photosensitive haloes. In detail, the fraction of neutral metals drops below 50 per cent at roughly 0.5R vir in both cases, but the photoresistant haloes are able to retain a significant component of neutral gas out to nearly the virial radius. Moreover, the total mass of circumgalactic metals around massive haloes grows owing to continued star formation, metal expulsion, and possibly stripping of enriched gas from infalling satellites, as can clearly be seen in Fig. 3. This means that, although photoresistant haloes are exposed to a generally stronger EUVB, their denser CGM are able to attenuate the ionization fronts and preserve a reservoir of neutral metals that extends throughout much ofthehaloevenatz = 6. In summary, Figs 7 and 8 suggest that the overall abundance of O I absorbers is regulated by a competition between the halo abundance, which grows in time, and absorption cross-section, which declines for all halo masses as time progresses. All haloes generate an enriched CGM down to the hydrogen-cooling limit. The metals remain largely neutral at z = 10 such that photosensitive haloes are the predominant source of low-ionization metal absorbers prior to reionization. Near the epoch of overlap, photosensitive haloes are completely ionized at radii larger than 0.5R vir whereas photoresistant haloes are more than 10 per cent neutral out to the virial radius. Hence, the typical host halo mass of O I absorbers increases as reionization proceeds. We will quantify this evolution in Figs 10 and MODELING OBSERVATIONS In this section, we relate the properties of individual haloes to volume-averaged statistical measurements of O I. Our analysis follows the approach adopted by many previous numerical studies of DLAs (e.g. Katz et al. 1996). We begin by computing the geometric cross-section for haloes to be observed in absorption in a way that relaxes the assumption of spherical symmetry. We then study how reionization affects the appearance of different haloes in absorption. Finally, we apply our cross-sections to predict the observable number density of absorbers and compare directly to observations. 4.1 Cross-section for observability Haloes that are more massive at a given redshift or at lower redshift for a given mass have produced more metals, leading to a higher cross-section. Similarly, haloes with lower mass at a given redshift should have a lower cross-section both because they have produced fewer metals and because they are more susceptible to an EUVB. Our simulations allow us to quantify these effects with minimal assumptions. We begin by computing the geometric cross-section for a halo to appear as an O I absorber with a column density greater than cm 2, which is the 50 per cent completeness limit reported by Becker et al. (2011). Computing the cross-sections accurately requires us to relax the assumption of spherical symmetry because the O I column density profiles are influenced by the filamentary structure of the gas density field (Fig. 3). We map each of our haloes on to a mesh with cells of width 200 physical pc including all gas out to twice the virial radius and then count the fraction of lines of sight passing within one virial radius for which the O I column density exceeds cm 2. We recompute this fraction using lines of sight in the x, y and z directions and average the three results. Using a finer mesh decreases Figure 9. The fraction of the area within one virial radius πrvir 2 that is covered by lines of sight with a neutral oxygen column greater than cm 2 as a function of halo mass at z = 10 (red crosses) and 6 (blue points). the cross-section while increasing the number of lines of sight with high columns, but the effect is weak; we have verified that using a mesh with twice this spatial resolution changes the cross-sections by 10 per cent. Considering lines of sight that pass outside of one virial radius would primarily have the effect of picking up absorption owing to neighbouring haloes, as can be seen at z = 6inFig.3. Incorporating the full three-dimensional gas distribution in this way automatically accounts for any departures from spherical symmetry. This means that, whereas we excluded gas that is closely associated with galaxies in Section 3.2 in order not to smear satellites over spherical shells when computing mean radial profiles, we include all halo gas in the analysis throughout the rest of this work. We show in Fig. 9 how the fraction of the area within one virial radius that is covered by observable lines of sight σ 14 /σ vir varies with halo mass at z = 10 (red crosses) and z = 6 (blue points). Broadly, σ 14 /σ vir < 1evenatz = 10. In detail, haloes more massive than 10 8 M are generally visible throughout much of the virial radius because the EUVB has not yet penetrated deep into the CGM. Haloes less massive than 10 8 M show weaker absorption because they are not capable of producing stars and metals even in a neutral IGM. At z = 6, the signature of reionization is obvious. The EUVB has penetrated well into the typical halo, suppressing the covered fraction to per cent. The threshold halo mass below which the absorption cross-section vanishes grows from 10 8 M at z = 10 to roughly M at z = 6. This owes to the combined effects of photoionization and gas exhaustion on photosensitive haloes. Razoumov et al. (2006) used radiation hydrodynamic simulations to find that haloes less massive than M retain the ability to accrete gas following reionization. Our simulations indicate a slightly higher threshold, likely owing to the tendency for outflows to reduce the gas density near halo cores. There is also a population of haloes at both redshifts that produce no observable absorption (σ 14 = 0). This population extends to higher mass at z = 6thanat z = 10, indicating that it is not purely an artefact of limited mass

10 O I-absorbing haloes 1827 resolution; instead, it reflects the weak star formation efficiencies and optically thin CGM of photosensitive haloes. 4.2 The dominant host haloes As a first application of our cross-sections, we may compute the most likely host halo mass for absorbers at a given column density. We do this by computing the probability density P(M N) thatan absorber with column density N O I > cm 2 is hosted by a halo of mass M<M h <M+ dm using Bayes theorem: P (M N) P (N M)P (M), (3) where P(N M) is the probability that a line of sight passing within the virial radius of a halo of mass M encounters a column greater than >10 14 cm 2 and P(M) is the prior probability of passing within a virial radius of a halo of mass M. The former is simply the ratio of the area within which the column exceeds cm 2 to the area within a virial radius (that is, σ 14 /σ vir ), and the latter is the fraction of haloes in this mass range weighted by the area within a virial radius. We show this probability density at z = 6 and 10 in Fig. 10. At z = 10, the distribution of halo masses that can host an observable system is weighted towards the hydrogen-cooling limit partly because the enriched CGM in such haloes remain mostly neutral, and partly because more massive haloes are not yet abundant enough to compete. By z = 6, the peak of the probability density function has shifted to higher mass by a factor of 2 3 because photosensitive haloes lose their gas while photoresistant haloes begin to assemble in force. Still, however, the characteristic host halo s mass lies within the range that is sensitive to photoionization heating (Finlator et al. 2011b). This suggests that, at any redshift, low-ionization metal absorbers probe the lowest mass haloes that retain the ability to form stars. Figure 10. The probability density that the host halo of an O I absorber with column density greater than cm 2 has a given mass at z = 10 and 6. O I absorbers are dominated by haloes a factor of less massive than the haloes that host Lyman-break galaxies and Lyα emitters (Ouchi et al. 2010; Muñoz & Loeb 2011). How do the host haloes of O I absorbers compare with the host haloes of galaxies that are selected in emission? Muñoz & Loeb (2011) used a detailed comparison between an analytic model and observations of Lyman-break galaxies at z = 7 8 to show that current observations likely do not probe below a halo mass of M. Similarly, Ouchi et al. (2010) have used clustering observations to infer that Lyα emitters live in haloes with masses between and M. Hence, absorption-selected samples trace star formation in haloes that are times less massive than the haloes that host emission-selected samples. This supports the suggestion by Becker et al. (2011) that studies in absorption offer more direct insight into the nature of the systems whose ionizing flux may have driven hydrogen reionization (see, e.g. Yan & Windhorst 2004; Alvarez et al. 2012; Robertson et al. 2013). Fig. 10 also gives insight into what would be required to observe the host galaxies of O I absorbers in emission. The typical O I absorber at z = 6 lives in a 10 9 M halo. Our models predict that the mean SFR of such haloes is M yr 1 (Finlator et al. 2011b). Assuming that the ratio of luminosity to SFR is ergs s 1 Hz 1 (M yr 1 ) 1 (Finlator, Oppenheimer & Davé 2011a, note that this includes an estimate for dust extinction), this corresponds to a rest-frame ultraviolet absolute magnitude of 14. This is roughly three magnitudes fainter than has been achieved at z 6 with the Hubble Space Telescope (Bouwens et al. 2012), and slightly fainter than will be achieved with the James Webb Space Telescope. As a caveat to Fig. 10, we note that the typical host halo mass at z = 6 may be underestimated because the most massive haloes are undersampled by our small simulation volume. In Fig. 12, we will use an analytic fit to our results to extrapolate to higher masses and confirm that photosensitive haloes still dominate. 4.3 The contribution of haloes of different masses Median cross-section versus mass By how much does the cross-section shrink from z = 10 6? Fig. 9 shows that the covered fraction declines at constant mass, but the x-axis is in units of virial radii. In practice, it is convenient to quantify the absorber abundance in terms of the number per absorption path length (Section 4.3.2), which in turn depends on the cross-section in proper units. To this end, we show in Fig. 11 how the median cross-section for observability σ 14 in proper kpc 2 depends on halo mass at six redshifts. The median trend evolves in three distinct ways. First, we still see a low-mass cutoff that grows owing to the gradual encroachment of ionization fronts into photosensitive haloes. As before, the cutoff evolves from <10 8 M at z = 10 to a few 10 8 M by z = 6. Secondly, for haloes with virial mass <10 10 M, the growth of the EUVB dominates over the impact of continuing metal enrichment with the result that the crosssection at a given halo mass shrinks. This is the signature of reionization: as observations probe higher redshifts, the EUVB weakens, haloes are more neutral, and absorption shifts from high-ionization transitions in photoresistant haloes to low-ionization transitions in photosensitive haloes. This is consistent with the observation that the abundance of C IV absorbers declines at z>6 while the abundance of low-ionization systems does not (Becker et al. 2011). The shift to lower host halo masses may manifest as a decline in the characteristic velocity width as observations push past z = 5; we will explore this possibility in future work. Finally, the cross-section for photoresistant haloes grows following z = 6 (that is, the solid green curve lies above the dotted blue one for z = 5andM h /M > ). This evolution is a direct response to our strongly redshift-dependent